QAE
Quantum Amplitude Estimation
Quantum kernel estimation and classification on NISQ backends: transparent, reproducible, and benchmarked on Superpositions Studio.
Overview
Quantum Amplitude Estimation (QAE) estimates the amplitude a of a marked subspace prepared by an operator A — equivalently, the probability of an event, an expected value, or a definite integral encoded into a quantum state. Classical Monte Carlo sampling estimates the same quantity to an additive error ε using O(1/ε²) samples, because sampling error only shrinks as 1/√N. QAE instead estimates a with O(1/ε) oracle calls to A: a quadratic reduction in the number of calls needed for the same precision.
Three variants are in common use. Canonical QAE, from the original Brassard–Høyer–Mosca–Tapp construction, reads out a via quantum phase estimation (QPE) and needs a wide, low-error ancilla register. Maximum-Likelihood QAE (MLAE) replaces QPE with a family of Grover-amplified circuits fit with a classical likelihood function. Iterative QAE (IQAE) narrows a confidence interval on a step by step, resetting the circuit before each round's chain of Grover applications accumulates too much error — which is why it is the variant used in the VaR/CVaR risk-estimation proof of concept below and in the app. See the full comparison against classical Monte Carlo sampling for where each method wins in practice.
In query-complexity terms: classical Monte Carlo needs N_MC(ε, α) = Θ(log(1/α) / ε²) samples to reach additive error ε at confidence 1-α, so its error shrinks only as 1/√N. IQAE needs N_IQAE(ε, α) = O(log(2/α · log²(π/4ε)) / ε) oracle calls for the same guarantee — asymptotically ε ~ 1/N instead of ε ~ 1/√N, up to a doubly-logarithmic factor. The two curves cross around ε ≈ 10⁻², and IQAE pulls further ahead of Monte Carlo as the required precision increases.
Why IQAE Matters on NISQ Hardware
IQAE does not depend on quantum phase estimation, so it needs neither the wide ancilla register nor the long coherence time canonical QAE requires. Because its Grover-style operator, borrowed from Grover's algorithm, is applied only a scheduled, limited number of times before the circuit resets, errors from decoherence and gate noise cannot build up across one unbounded chain of applications. That is what makes IQAE a practical amplitude-estimation algorithm today, rather than only an asymptotic theoretical construct.
How QAE Works
The iterative loop that turns a quantum amplitude into a bounded numerical estimate
Prepare the distribution
Build operator A = LU: U embeds a discretized probability distribution (for example, asset log-returns sampled at 2ⁿ points) into the state, and oracle L flips an ancilla qubit for outcomes past a loss threshold.
Build the Grover-like operator
Construct Q = -AS₀A†Sχ from reflections about the zero state and the marked subspace — the same reflection structure Grover's algorithm uses to amplify a marked outcome, repurposed here to rotate the amplitude instead of a search index.
Amplify and estimate iteratively
Apply Qᵏ for a scheduled sequence of k values, measuring the ancilla each round and counting successes. Iterative QAE narrows the estimate round by round instead of committing to one long coherent phase-estimation circuit.
Track a confidence interval
After each round, turn the observed hit counts into a Chernoff-Hoeffding or Clopper-Pearson confidence interval [θ_l, θ_u] on the amplitude's angle, and choose the next k that keeps the interval shrinking unambiguously.
Read out the result
Convert the final estimate of a = sin²θₐ into the requested quantity — a tail probability for VaR, a conditional expectation for CVaR, an option price, or an integral — to within target error ε at confidence 1-α.
Real-World Applications
Anything expressible as an amplitude of a quantum state is a candidate for QAE
Because IQAE returns an estimate with a guaranteed confidence interval rather than a single noisy sample, it fits naturally anywhere a business process already tolerates a Monte Carlo estimate and would benefit from tightening that estimate's error bars for a given compute budget.
Financial Risk (VaR & CVaR)
Estimate the probability that portfolio losses exceed a threshold, and the expected loss beyond it, from a quantum-encoded return distribution
Option Pricing
Price options by estimating the expected discounted payoff as an amplitude — one of the earliest proposed applications of amplitude estimation
Expectation Values
Estimate the expectation value of an observable encoded into a quantum state, with the same quadratic query advantage over sampling
Numerical Integration
Approximate a definite integral by encoding the integrand as an amplitude and reading it out with fewer oracle calls than classical sampling-based quadrature
The worked example on this page is portfolio risk: our VaR/CVaR risk-estimation proof of concept uses IQAE on real AAPL and MSFT price data. For the query-count math behind the speedup, see the QAE vs. Monte Carlo comparison.
Strengths & Limitations
Strengths
- Quadratic speedup over classical Monte Carlo: O(1/ε) oracle calls versus O(1/ε²) samples for the same additive error ε, a gap that widens as the required precision increases
- IQAE avoids quantum phase estimation entirely, so it needs neither the wide ancilla register nor the long coherence time canonical QAE requires to read out a precise phase
- The iterative schedule applies the Grover operator a limited, scheduled number of times before resetting, capping how much decoherence and gate noise can build up in a single estimate
- Amplitude estimation still returns a valid, bounded confidence interval even when only short Grover-operator sequences are available on noisy hardware, rather than failing outright
Limitations
- Building the loss/payoff oracle from elementary gates is nontrivial: the proof of concept's distribution-embedding and loss operators together needed 500–1,000 gates for two assets
- Noise lowers the precision of the amplitude bounds on the later, higher-k rounds of the iteration — exactly where the tightest confidence intervals matter most
- Currently practical only for a limited number of assets; classical simulation of the full circuit for verification was capped at 16 qubits for computational cost reasons
- The full quadratic speedup only pays off at industrial scale on fault-tolerant hardware — public roadmaps place that generation of devices in the 2030s
Benchmarking & Verification
Compare IQAE's estimate against a classical Monte Carlo baseline run on the same distribution; report MSE, relative error, run count and shots per run. The risk-estimation proof of concept measured a 0.001 VaR MSE and a 0.006 CVaR MSE against Monte Carlo on 11 qubits — see the full QAE vs. Monte Carlo comparison for the query-count math behind that result.
Hardware & Requirements
Note: fault-tolerant, low-error hardware is what turns IQAE's asymptotic advantage into a working one. Resource estimates for industrial-scale derivative pricing put that at thousands of logical qubits and roughly 10⁷–10⁹ logical gate operations — public hardware roadmaps place that generation of devices in the 2030s.
Proof-of-Concept
Real experimental results from an IQAE VaR/CVaR run on live market data
The results below come from a full IQAE run on real market data, not a synthetic toy distribution: two large-cap stocks, discretized into a return distribution and fed through the state-preparation and loss operators described in How QAE Works above, then compared against a classical Monte Carlo baseline on the same distribution.
Task
VaR and CVaR estimation on AAPL and MSFT log-price time series, discretized and embedded as a probability distribution
Dataset
9,909 time steps across 2 assets, converted to log-scale growth rates and modeled as a Gaussian process
11
Qubits used (5/asset × 2 + 1 ancilla)
1,000
Shots per run
0.001
VaR estimation MSE vs. Monte Carlo
0.006
CVaR estimation MSE vs. Monte Carlo
202 ms
Runtime per run
20.6 s
Total execution time (simulator)
These results come from the full VaR/CVaR risk-estimation proof of concept, which also walks through the dataset preparation and the quantum-vs-classical comparison in detail.
FAQ
Common questions about Quantum Amplitude Estimation and IQAE
Is QAE faster than Monte Carlo?
Yes, in query count: QAE needs O(1/ε) oracle calls to reach additive error ε, while classical Monte Carlo sampling needs O(1/ε²) samples for the same error — a quadratic reduction in calls. See the QAE vs. Monte Carlo comparison for where classical sampling still wins in practice today.
What is Iterative Quantum Amplitude Estimation (IQAE)?
IQAE is a QAE variant that narrows a confidence interval on the target amplitude step by step, applying the Grover operator a scheduled number of times each round and resetting the circuit before errors build up, instead of running one long phase-estimation circuit like canonical QAE.
How many qubits does QAE need?
It depends on the distribution's resolution. The risk-estimation proof of concept on this page used 5 qubits per asset plus 1 ancilla — 11 qubits for a 2-asset portfolio; a 100-asset portfolio would need roughly 700 qubits at that resolution.
Does QAE need quantum phase estimation?
Canonical QAE does, which is why it needs a wide, low-error ancilla register. IQAE and Maximum-Likelihood QAE were designed specifically to avoid QPE, which is why IQAE is the variant used in near-term implementations, including the one on this page.
What can QAE estimate besides Value at Risk?
Anything expressible as the amplitude of a quantum state: expected values, option prices, and definite integrals, alongside the VaR/CVaR tail-probability estimation covered in this page's proof of concept.
Can QAE run on today's hardware?
IQAE runs on simulators and small NISQ devices today; the proof of concept here ran on an 11-qubit simulator. A clear industrial-scale advantage over classical Monte Carlo for realistic financial models most likely needs fault-tolerant hardware, which public roadmaps place in the 2030s.
Related pages
Algorithms, applied use cases and benchmarks connected to this page
Finance
VaR and CVaR Estimation with IQAE
Estimate Value at Risk and Conditional Value at Risk with Iterative Quantum Amplitude Estimation, Monte Carlo benchmarks, and reproducible reports.
Comparisons
QAE vs Monte Carlo for Risk Estimation
Compare quantum amplitude estimation (IQAE) with classical Monte Carlo for VaR/CVaR risk estimation, an 11-qubit benchmark, and when each wins.
Algorithms
Grover's Algorithm for Quantum Search
Learn and benchmark Grover's algorithm for quantum search with reproducible circuits, oracle constructions, and implementation notes.
Algorithms
Quantum Reservoir Computing (QRC)
Fixed quantum reservoir plus a trained linear read-out for time-series forecasting, benchmarked against Ridge, ESN, and LSTM on real S&P 500 volatility data.
Ready to Run QAE?
Estimate probabilities, expectations, and portfolio risk with a quadratic speedup over Monte Carlo, and export reproducible code and comprehensive reports.
Powered by Superpositions Studio: transparent, reproducible quantum computing